It is 2 issues. One has to do with any details (faces are only psychologically more harder in impact, and hands are complex), and one has to do with topology. 4 or 6 fingers on a big hand in closeup is topology related.
I see systems like DallE as universal Super-Morphers. They can blend everything in to everything else, this is a main power of this systems. And so they can put completely convincingly a other finger on a hand. (but i write here on the wrong place…)
PS anyway… This is probably the reason why they use facial templates (over training), they not trust there systems 100%. And distorted faces are very confusing for most people.
It could be that the developers are now switching of some “templates”, so we will see some more errors if this is so. sadly the developers not communicate there doings.